Update README.md
Browse files
README.md
CHANGED
|
@@ -31,6 +31,8 @@ Our framework unifies three stages into an end-to-end pipeline:
|
|
| 31 |
- Model Training — a memory-augmented Diffusion Transformer (DiT) with an error buffer that learns action-conditioned generation with memory-enhanced long-horizon consistency;
|
| 32 |
- Inference Deployment — few-step sampling, INT8 quantization, and model distillation achieving 720p@40FPS real-time generation with a 5B model.
|
| 33 |
|
|
|
|
|
|
|
| 34 |
## ✨ Key Features
|
| 35 |
- 🚀 **Feature 1**: **Upgraded Data Engine**: Combines Unreal Engine-based synthetic data, large-scale automated AAA game data, and real-world video augmentation to generate high-quality Video–Pose–Action–Prompt data.
|
| 36 |
- 🖱️ **Feature 2**: **Long-horizon Memory & Consistency**: Uses prediction residuals and frame re-injection for self-correction, while camera-aware memory ensures long-term spatiotemporal consistency.
|
|
|
|
| 31 |
- Model Training — a memory-augmented Diffusion Transformer (DiT) with an error buffer that learns action-conditioned generation with memory-enhanced long-horizon consistency;
|
| 32 |
- Inference Deployment — few-step sampling, INT8 quantization, and model distillation achieving 720p@40FPS real-time generation with a 5B model.
|
| 33 |
|
| 34 |
+
|
| 35 |
+
|
| 36 |
## ✨ Key Features
|
| 37 |
- 🚀 **Feature 1**: **Upgraded Data Engine**: Combines Unreal Engine-based synthetic data, large-scale automated AAA game data, and real-world video augmentation to generate high-quality Video–Pose–Action–Prompt data.
|
| 38 |
- 🖱️ **Feature 2**: **Long-horizon Memory & Consistency**: Uses prediction residuals and frame re-injection for self-correction, while camera-aware memory ensures long-term spatiotemporal consistency.
|